Skip to contentSkip to conversation

5.6-5.10 balance regression?

DISCUSSION Stéphane Lesimple 5 participants 13 messages Dec 27, 2020 – Dec 29, 2020 Btrfs file system

lore
Share & tools
lore
#0 lore raw

Hello,

As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.

After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:

Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1
Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents
Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2

Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:

Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start -ddevid=5,limit=1
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1
Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents
Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2

There were of course still plenty of room on the FS, as I added a new 16T drive (a btrfs fi usage is further down this email), so it struck me as odd.
So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to complete immediately given that there were already a copy of this data present on another drive:

Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1
Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1
Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents
Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2

That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing, using btrfs dev del missing, but the balance still failed with the same error on the same block group.

So, as I was running 5.10.1 just for a few days, I tried an older kernel: 5.6.17, and retried the balance once again (with still the drive voluntarily missing):

[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts
[ 413.188814] BTRFS info (device dm-10): using free space tree
[ 413.188815] BTRFS info (device dm-10): has skinny extents
[ 413.189674] BTRFS warning (device dm-10): devid 5 uuid 068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing
[ 424.159486] BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1
[ 424.772640] BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1
[ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage: move data extents
[ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage: update data pointers
[ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0

The problematic block group was balanced successfully this time.

I balanced a few more successfully (without the -dconvert=single option), then decided to reboot under 5.10 just to see if I would hit this issue again. I didn't: the btrfs dev del worked correctly after the last 500G or so data was moved out of the drive.

This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":

Overall:

Device size:        41.89TiB
Device allocated:   21.74TiB
Device unallocated: 20.14TiB
Device missing:      9.09TiB
Used:               21.71TiB
Free (estimated):   10.08TiB (min: 10.07TiB)
Data ratio:             2.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:       yes (data)

Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T 3.00GiB

Data,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta 7.14TiB
/dev/mapper/luks-10Tb 7.10TiB
missing 482.00GiB
/dev/mapper/luks-new16T 6.95TiB

Metadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%)
/dev/mapper/luks-10Tb 36.00GiB
/dev/mapper/luks-ssd-mdata 36.00GiB

System,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta 32.00MiB
/dev/mapper/luks-10Tb 32.00MiB

Unallocated:

/dev/mapper/luks-10Ta       1.95TiB
/dev/mapper/luks-10Tb       1.96TiB
missing                     8.62TiB

/dev/mapper/luks-ssd-mdata 11.29GiB
/dev/mapper/luks-new16T 7.60TiB

I wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping into the same issue on another FS, while rebalancing data after adding a drive, which happens to be the old 10T drive of the FS above.

The btrfs fi usage of this second FS is as follows:

Overall:

Device size:        25.50TiB
Device allocated:   22.95TiB
Device unallocated:  2.55TiB
Device missing:        0.00B
Used:               22.36TiB
Free (estimated):    3.14TiB (min: 1.87TiB)
Data ratio:             1.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:        no

Data,single: Size:22.89TiB, Used:22.29TiB (97.40%)
/dev/mapper/luks-12T 10.91TiB

/dev/mapper/luks-3Ta         2.73TiB
/dev/mapper/luks-3Tb         2.73TiB
/dev/mapper/luks-10T         6.52TiB

Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%) /dev/mapper/luks-ssd-mdata2 32.00GiB
/dev/mapper/luks-10T 32.00GiB

System,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb 32.00MiB
/dev/mapper/luks-10T 32.00MiB

Unallocated:
/dev/mapper/luks-12T 45.00MiB

/dev/mapper/luks-ssd-mdata2  4.00GiB
/dev/mapper/luks-3Ta         1.02MiB
/dev/mapper/luks-3Tb         2.97GiB
/dev/mapper/luks-10T         2.54TiB

I can reproduce the problem reliably:

# btrfs bal start -dvrange=34625344765952..34625344765953 /tank
ERROR: error during balancing '/tank': No such file or directory
There may be more info in syslog - try dmesg | tail
[145979.563045] BTRFS info (device dm-10): balance: start -dvrange=34625344765952..34625344765953
[145979.585572] BTRFS info (device dm-10): relocating block group 34625344765952 flags data|raid1
[145990.396585] BTRFS info (device dm-10): found 167 extents, stage: move data extents
[146002.236115] BTRFS info (device dm-10): balance: ended with status: -2

If anybody is interested in looking into this, this time I can leave the FS in this state. The issue is reproducible, and I can live without completing the balance for the next weeks or even months, as I don't think I'll need the currently unallocatable space soon.

I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).

I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave like this, but on the first success, I won't know how to reproduce the issue a second time, as I'm not 100% sure it can be done solely with the btrfs-image.

Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.html

Regards,

Stéphane Lesimple.

#1 lore raw

Hi,

last week I had the same problem on a btrfs filesystem after updating to kernel 5.10.1. I have never had this problem before kernel 5.10.x. 5.9.x did now show any problem.

Dec 14 22:30:59 xxx kernel: BTRFS info (device sda2): scrub: started on devid 1
Dec 14 22:31:09 xxx kernel: BTRFS info (device sda2): scrub: finished on devid 1 with status: 0
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: start -dusage=10
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): relocating block group 71694286848 flags data
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): found 1058 extents, stage: move data extents
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: ended with status: -2

This is not a multidevice volume but a volume consisting of a single partition.

xxx ~ # btrfs fi df /u00
Data, single: total=10.01GiB, used=9.24GiB
System, single: total=4.00MiB, used=16.00KiB
Metadata, single: total=2.76GiB, used=1.10GiB
GlobalReserve, single: total=47.17MiB, used=0.00B

xxx ~ # btrfs device usage /u00
/dev/sda2, ID: 1
   Device size:            19.81GiB
   Device slack:              0.00B
   Data,single:            10.01GiB
   Metadata,single:         2.76GiB
   System,single:           4.00MiB
   Unallocated:             7.04GiB

Quoted 169 lines · Stéphane Lesimple

Hello,

As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.

After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:

Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1
Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents
Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2

Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:

Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start -ddevid=5,limit=1
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1
Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents
Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2

There were of course still plenty of room on the FS, as I added a new 16T drive (a btrfs fi usage is further down this email), so it struck me as odd.
So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to complete immediately given that there were already a copy of this data present on another drive:

Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1
Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1
Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents
Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2

That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing, using btrfs dev del missing, but the balance still failed with the same error on the same block group.

So, as I was running 5.10.1 just for a few days, I tried an older kernel: 5.6.17, and retried the balance once again (with still the drive voluntarily missing):

[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts
[ 413.188814] BTRFS info (device dm-10): using free space tree
[ 413.188815] BTRFS info (device dm-10): has skinny extents
[ 413.189674] BTRFS warning (device dm-10): devid 5 uuid 068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing
[ 424.159486] BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1
[ 424.772640] BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1
[ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage: move data extents
[ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage: update data pointers
[ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0

The problematic block group was balanced successfully this time.

I balanced a few more successfully (without the -dconvert=single option), then decided to reboot under 5.10 just to see if I would hit this issue again. I didn't: the btrfs dev del worked correctly after the last 500G or so data was moved out of the drive.

This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":

Overall:

Device size:        41.89TiB
Device allocated:   21.74TiB
Device unallocated: 20.14TiB
Device missing:      9.09TiB
Used:               21.71TiB
Free (estimated):   10.08TiB (min: 10.07TiB)
Data ratio:             2.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:       yes (data)

Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T 3.00GiB

Data,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta 7.14TiB
/dev/mapper/luks-10Tb 7.10TiB
missing 482.00GiB
/dev/mapper/luks-new16T 6.95TiB

Metadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%)
/dev/mapper/luks-10Tb 36.00GiB
/dev/mapper/luks-ssd-mdata 36.00GiB

System,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta 32.00MiB
/dev/mapper/luks-10Tb 32.00MiB

Unallocated:

/dev/mapper/luks-10Ta       1.95TiB
/dev/mapper/luks-10Tb       1.96TiB
missing                     8.62TiB

/dev/mapper/luks-ssd-mdata 11.29GiB
/dev/mapper/luks-new16T 7.60TiB

I wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping into the same issue on another FS, while rebalancing data after adding a drive, which happens to be the old 10T drive of the FS above.

The btrfs fi usage of this second FS is as follows:

Overall:

Device size:        25.50TiB
Device allocated:   22.95TiB
Device unallocated:  2.55TiB
Device missing:        0.00B
Used:               22.36TiB
Free (estimated):    3.14TiB (min: 1.87TiB)
Data ratio:             1.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:        no

Data,single: Size:22.89TiB, Used:22.29TiB (97.40%)
/dev/mapper/luks-12T 10.91TiB

/dev/mapper/luks-3Ta         2.73TiB
/dev/mapper/luks-3Tb         2.73TiB
/dev/mapper/luks-10T         6.52TiB

Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%) /dev/mapper/luks-ssd-mdata2 32.00GiB
/dev/mapper/luks-10T 32.00GiB

System,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb 32.00MiB
/dev/mapper/luks-10T 32.00MiB

Unallocated:
/dev/mapper/luks-12T 45.00MiB

/dev/mapper/luks-ssd-mdata2  4.00GiB
/dev/mapper/luks-3Ta         1.02MiB
/dev/mapper/luks-3Tb         2.97GiB
/dev/mapper/luks-10T         2.54TiB

I can reproduce the problem reliably:

# btrfs bal start -dvrange=34625344765952..34625344765953 /tank
ERROR: error during balancing '/tank': No such file or directory
There may be more info in syslog - try dmesg | tail
[145979.563045] BTRFS info (device dm-10): balance: start -dvrange=34625344765952..34625344765953
[145979.585572] BTRFS info (device dm-10): relocating block group 34625344765952 flags data|raid1
[145990.396585] BTRFS info (device dm-10): found 167 extents, stage: move data extents
[146002.236115] BTRFS info (device dm-10): balance: ended with status: -2

If anybody is interested in looking into this, this time I can leave the FS in this state. The issue is reproducible, and I can live without completing the balance for the next weeks or even months, as I don't think I'll need the currently unallocatable space soon.

I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).

I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave like this, but on the first success, I won't know how to reproduce the issue a second time, as I'm not 100% sure it can be done solely with the btrfs-image.

Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.html

Regards,

Stéphane Lesimple.

#2 lore raw
Quoted 28 lines · David Arendt

Hi,

last week I had the same problem on a btrfs filesystem after updating to kernel 5.10.1. I have never had this problem before kernel 5.10.x. 5.9.x did now show any problem.

Dec 14 22:30:59 xxx kernel: BTRFS info (device sda2): scrub: started on
devid 1
Dec 14 22:31:09 xxx kernel: BTRFS info (device sda2): scrub: finished on
devid 1 with status: 0
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: start
-dusage=10
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): relocating block
group 71694286848 flags data
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): found 1058
extents, stage: move data extents
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: ended

with status: -2

This is not a multidevice volume but a volume consisting of a single partition.

xxx ~ # btrfs fi df /u00
Data, single: total=10.01GiB, used=9.24GiB
System, single: total=4.00MiB, used=16.00KiB
Metadata, single: total=2.76GiB, used=1.10GiB
GlobalReserve, single: total=47.17MiB, used=0.00B

Quoted 8 lines

xxx ~ # btrfs device usage /u00
/dev/sda2, ID: 1
   Device size:            19.81GiB
   Device slack:              0.00B
   Data,single:            10.01GiB
   Metadata,single:         2.76GiB
   System,single:           4.00MiB
   Unallocated:             7.04GiB

This seems small enough, thus a btrfs-image dump would help.

Although there is a limit for btrfs-image dump, since it only contains metadata, when we try to balance data to reproduce the bug, it would easily cause data csum error and exit convert.

If possible, would you please try to take a dump with this branch?
https://github.com/adam900710/btrfs-progs/tree/image_data_dump

It provides a new option for btrfs-image, -d, which will also take the data.

Also, please keep in mind that, -d dump will contain data of your fs, thus if it contains confidential info, please use regular btrfs-image.

Thanks,
Qu

Quoted 212 lines · Stéphane Lesimple
Quoted 207 lines

Hello,

As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process
of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.

After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:

Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:

Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start
-ddevid=5,limit=1
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

There were of course still plenty of room on the FS, as I added a new 16T drive
(a btrfs fi usage is further down this email), so it struck me as odd. So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to
complete immediately given that there were already a copy of this data present on another drive:

Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start
-dconvert=single,soft,devid=5,limit=1
Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing,
using btrfs dev del missing, but the balance still failed with the same error on the same block group.

So, as I was running 5.10.1 just for a few days, I tried an older
kernel: 5.6.17,
and retried the balance once again (with still the drive voluntarily missing):

[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts
[ 413.188814] BTRFS info (device dm-10): using free space tree
[ 413.188815] BTRFS info (device dm-10): has skinny extents
[ 413.189674] BTRFS warning (device dm-10): devid 5 uuid
068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing
[ 424.159486] BTRFS info (device dm-10): balance: start
-dconvert=single,soft,devid=5,limit=1
[ 424.772640] BTRFS info (device dm-10): relocating block group
11115169841152 flags data|raid1
[ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage:
move data extents
[ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage:
update data pointers
[ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0

The problematic block group was balanced successfully this time.

I balanced a few more successfully (without the -dconvert=single option), then decided to reboot under 5.10 just to see if I would hit this issue again.
I didn't: the btrfs dev del worked correctly after the last 500G or so data
was moved out of the drive.

This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":

Overall:
Device size:        41.89TiB
Device allocated:   21.74TiB
Device unallocated: 20.14TiB
Device missing:      9.09TiB
Used:               21.71TiB
Free (estimated):   10.08TiB (min: 10.07TiB)
Data ratio:             2.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:       yes (data)

Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T     3.00GiB

Data,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta       7.14TiB
/dev/mapper/luks-10Tb       7.10TiB
missing                   482.00GiB
/dev/mapper/luks-new16T     6.95TiB

Metadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%) /dev/mapper/luks-10Tb      36.00GiB /dev/mapper/luks-ssd-mdata 36.00GiB

System,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta      32.00MiB
/dev/mapper/luks-10Tb      32.00MiB

Unallocated:
/dev/mapper/luks-10Ta       1.95TiB
/dev/mapper/luks-10Tb       1.96TiB
missing                     8.62TiB
/dev/mapper/luks-ssd-mdata 11.29GiB
/dev/mapper/luks-new16T     7.60TiB

I wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping
into the same issue on another FS, while rebalancing data after adding a drive,
which happens to be the old 10T drive of the FS above.

The btrfs fi usage of this second FS is as follows:

Overall:
Device size:        25.50TiB
Device allocated:   22.95TiB
Device unallocated:  2.55TiB
Device missing:        0.00B
Used:               22.36TiB
Free (estimated):    3.14TiB (min: 1.87TiB)
Data ratio:             1.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:        no

Data,single: Size:22.89TiB, Used:22.29TiB (97.40%) /dev/mapper/luks-12T        10.91TiB /dev/mapper/luks-3Ta         2.73TiB /dev/mapper/luks-3Tb         2.73TiB /dev/mapper/luks-10T         6.52TiB

Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%) /dev/mapper/luks-ssd-mdata2 32.00GiB /dev/mapper/luks-10T        32.00GiB

System,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb        32.00MiB
/dev/mapper/luks-10T        32.00MiB

Unallocated:
/dev/mapper/luks-12T        45.00MiB
/dev/mapper/luks-ssd-mdata2  4.00GiB
/dev/mapper/luks-3Ta         1.02MiB
/dev/mapper/luks-3Tb         2.97GiB
/dev/mapper/luks-10T         2.54TiB

I can reproduce the problem reliably:

# btrfs bal start -dvrange=34625344765952..34625344765953 /tank
ERROR: error during balancing '/tank': No such file or directory
There may be more info in syslog - try dmesg | tail
[145979.563045] BTRFS info (device dm-10): balance: start
-dvrange=34625344765952..34625344765953
[145979.585572] BTRFS info (device dm-10): relocating block group
34625344765952 flags data|raid1
[145990.396585] BTRFS info (device dm-10): found 167 extents, stage:
move data extents
[146002.236115] BTRFS info (device dm-10): balance: ended with status: -2

If anybody is interested in looking into this, this time I can leave the FS in this state.
The issue is reproducible, and I can live without completing the balance for the next weeks
or even months, as I don't think I'll need the currently unallocatable space soon.

I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).

I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave
like this, but on the first success, I won't know how to reproduce the issue a second time, as
I'm not 100% sure it can be done solely with the btrfs-image.

Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.html

Regards,

Stéphane Lesimple.

#3 lore raw

Hi,

unfortunately the problem is no longer reproducible, probably due to writes happening in meantime. If you still want a btrfs-image, I can create one (unfortunately only without data as there is confidential data in it), but as the problem is currently no longer reproducible, I think it probably won't help.

Thanks in advance,
David Arendt

Quoted 268 lines · Qu Wenruo
Quoted 28 lines · David Arendt

Hi,

last week I had the same problem on a btrfs filesystem after updating to kernel 5.10.1. I have never had this problem before kernel 5.10.x. 5.9.x did now show any problem.

Dec 14 22:30:59 xxx kernel: BTRFS info (device sda2): scrub: started on
devid 1
Dec 14 22:31:09 xxx kernel: BTRFS info (device sda2): scrub: finished on
devid 1 with status: 0
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: start
-dusage=10
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): relocating block
group 71694286848 flags data
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): found 1058
extents, stage: move data extents
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: ended

with status: -2

This is not a multidevice volume but a volume consisting of a single partition.

xxx ~ # btrfs fi df /u00
Data, single: total=10.01GiB, used=9.24GiB
System, single: total=4.00MiB, used=16.00KiB
Metadata, single: total=2.76GiB, used=1.10GiB
GlobalReserve, single: total=47.17MiB, used=0.00B

Quoted 8 lines

xxx ~ # btrfs device usage /u00
/dev/sda2, ID: 1
    Device size:            19.81GiB
    Device slack:              0.00B
    Data,single:            10.01GiB
    Metadata,single:         2.76GiB
    System,single:           4.00MiB
    Unallocated:             7.04GiB

This seems small enough, thus a btrfs-image dump would help.

Although there is a limit for btrfs-image dump, since it only contains metadata, when we try to balance data to reproduce the bug, it would easily cause data csum error and exit convert.

If possible, would you please try to take a dump with this branch?
https://github.com/adam900710/btrfs-progs/tree/image_data_dump

It provides a new option for btrfs-image, -d, which will also take the data.

Also, please keep in mind that, -d dump will contain data of your fs, thus if it contains confidential info, please use regular btrfs-image.

Thanks,
Qu

Quoted 212 lines · Stéphane Lesimple
Quoted 207 lines

Hello,

As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process
of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.

After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:

Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:

Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start
-ddevid=5,limit=1
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

There were of course still plenty of room on the FS, as I added a new 16T drive
(a btrfs fi usage is further down this email), so it struck me as odd. So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to
complete immediately given that there were already a copy of this data present on another drive:

Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start
-dconvert=single,soft,devid=5,limit=1
Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing,
using btrfs dev del missing, but the balance still failed with the same error on the same block group.

So, as I was running 5.10.1 just for a few days, I tried an older
kernel: 5.6.17,
and retried the balance once again (with still the drive voluntarily missing):

[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts
[ 413.188814] BTRFS info (device dm-10): using free space tree
[ 413.188815] BTRFS info (device dm-10): has skinny extents
[ 413.189674] BTRFS warning (device dm-10): devid 5 uuid
068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing
[ 424.159486] BTRFS info (device dm-10): balance: start
-dconvert=single,soft,devid=5,limit=1
[ 424.772640] BTRFS info (device dm-10): relocating block group
11115169841152 flags data|raid1
[ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage:
move data extents
[ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage:
update data pointers
[ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0

The problematic block group was balanced successfully this time.

I balanced a few more successfully (without the -dconvert=single option), then decided to reboot under 5.10 just to see if I would hit this issue again.
I didn't: the btrfs dev del worked correctly after the last 500G or so data
was moved out of the drive.

This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":

Overall:
Device size:        41.89TiB
Device allocated:   21.74TiB
Device unallocated: 20.14TiB
Device missing:      9.09TiB
Used:               21.71TiB
Free (estimated):   10.08TiB (min: 10.07TiB)
Data ratio:             2.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:       yes (data)

Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T     3.00GiB

Data,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta       7.14TiB
/dev/mapper/luks-10Tb       7.10TiB
missing                   482.00GiB
/dev/mapper/luks-new16T     6.95TiB

Metadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%) /dev/mapper/luks-10Tb      36.00GiB /dev/mapper/luks-ssd-mdata 36.00GiB

System,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta      32.00MiB
/dev/mapper/luks-10Tb      32.00MiB

Unallocated:
/dev/mapper/luks-10Ta       1.95TiB
/dev/mapper/luks-10Tb       1.96TiB
missing                     8.62TiB
/dev/mapper/luks-ssd-mdata 11.29GiB
/dev/mapper/luks-new16T     7.60TiB

I wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping
into the same issue on another FS, while rebalancing data after adding a drive,
which happens to be the old 10T drive of the FS above.

The btrfs fi usage of this second FS is as follows:

Overall:
Device size:        25.50TiB
Device allocated:   22.95TiB
Device unallocated:  2.55TiB
Device missing:        0.00B
Used:               22.36TiB
Free (estimated):    3.14TiB (min: 1.87TiB)
Data ratio:             1.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:        no

Data,single: Size:22.89TiB, Used:22.29TiB (97.40%) /dev/mapper/luks-12T        10.91TiB /dev/mapper/luks-3Ta         2.73TiB /dev/mapper/luks-3Tb         2.73TiB /dev/mapper/luks-10T         6.52TiB

Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%) /dev/mapper/luks-ssd-mdata2 32.00GiB /dev/mapper/luks-10T        32.00GiB

System,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb        32.00MiB
/dev/mapper/luks-10T        32.00MiB

Unallocated:
/dev/mapper/luks-12T        45.00MiB
/dev/mapper/luks-ssd-mdata2  4.00GiB
/dev/mapper/luks-3Ta         1.02MiB
/dev/mapper/luks-3Tb         2.97GiB
/dev/mapper/luks-10T         2.54TiB

I can reproduce the problem reliably:

# btrfs bal start -dvrange=34625344765952..34625344765953 /tank
ERROR: error during balancing '/tank': No such file or directory
There may be more info in syslog - try dmesg | tail
[145979.563045] BTRFS info (device dm-10): balance: start
-dvrange=34625344765952..34625344765953
[145979.585572] BTRFS info (device dm-10): relocating block group
34625344765952 flags data|raid1
[145990.396585] BTRFS info (device dm-10): found 167 extents, stage:
move data extents
[146002.236115] BTRFS info (device dm-10): balance: ended with status: -2

If anybody is interested in looking into this, this time I can leave the FS in this state.
The issue is reproducible, and I can live without completing the balance for the next weeks
or even months, as I don't think I'll need the currently unallocatable space soon.

I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).

I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave
like this, but on the first success, I won't know how to reproduce the issue a second time, as
I'm not 100% sure it can be done solely with the btrfs-image.

Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.html

Regards,

Stéphane Lesimple.

#4 lore raw
Quoted 3 lines · David Arendt

Hi,

unfortunately the problem is no longer reproducible, probably due to writes happening in meantime. If you still want a btrfs-image, I can create one (unfortunately only without data as there is confidential data in it), but as the problem is currently no longer reproducible, I think it probably won't help.

That's fine, at least you get your fs back to normal.

I tried several small balance locally, not reproduced, thus I guess it may be related to certain tree layout.

Anyway, I'll wait for another small enough and reproducible report.

Thanks,
Qu

Quoted 274 lines

Thanks in advance,
David Arendt

Quoted 268 lines · Qu Wenruo
Quoted 28 lines · David Arendt

Hi,

last week I had the same problem on a btrfs filesystem after updating to kernel 5.10.1. I have never had this problem before kernel 5.10.x. 5.9.x did now show any problem.

Dec 14 22:30:59 xxx kernel: BTRFS info (device sda2): scrub: started on
devid 1
Dec 14 22:31:09 xxx kernel: BTRFS info (device sda2): scrub: finished on
devid 1 with status: 0
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: start
-dusage
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): relocating block
group 71694286848 flags data
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): found 1058
extents, stage: move data extents
Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: ended

with status: -2

This is not a multidevice volume but a volume consisting of a single partition.

xxx ~ # btrfs fi df /u00
Data, single: total.01GiB, used=24GiB
System, single: total=00MiB, used.00KiB
Metadata, single: total=76GiB, used=10GiB
GlobalReserve, single: totalG.17MiB, used=00B

Quoted 8 lines

xxx ~ # btrfs device usage /u00
/dev/sda2, ID: 1
    Device size:            19.81GiB
    Device slack:              0.00B
    Data,single:            10.01GiB
    Metadata,single:         2.76GiB
    System,single:           4.00MiB
    Unallocated:             7.04GiB

This seems small enough, thus a btrfs-image dump would help.

Although there is a limit for btrfs-image dump, since it only contains metadata, when we try to balance data to reproduce the bug, it would easily cause data csum error and exit convert.

If possible, would you please try to take a dump with this branch?
https://github.com/adam900710/btrfs-progs/tree/image_data_dump

It provides a new option for btrfs-image, -d, which will also take the data.

Also, please keep in mind that, -d dump will contain data of your fs, thus if it contains confidential info, please use regular btrfs-image.

Thanks,
Qu

Quoted 212 lines · Stéphane Lesimple
Quoted 207 lines

Hello,

As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process
of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.

After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:

Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:

Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start
-ddevid=limit=
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

There were of course still plenty of room on the FS, as I added a new 16T drive
(a btrfs fi usage is further down this email), so it struck me as odd. So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to
complete immediately given that there were already a copy of this data present on another drive:

Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start
-dconvert=ngle,soft,devid=limit=
Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating
block group 11115169841152 flags data|raid1
Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264
extents, stage: move data extents
Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: ended

with status: -2

That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing,
using btrfs dev del missing, but the balance still failed with the same error on the same block group.

So, as I was running 5.10.1 just for a few days, I tried an older
kernel: 5.6.17,
and retried the balance once again (with still the drive voluntarily missing):

[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts
[ 413.188814] BTRFS info (device dm-10): using free space tree
[ 413.188815] BTRFS info (device dm-10): has skinny extents
[ 413.189674] BTRFS warning (device dm-10): devid 5 uuid
068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing
[ 424.159486] BTRFS info (device dm-10): balance: start
-dconvert=ngle,soft,devid=limit=
[ 424.772640] BTRFS info (device dm-10): relocating block group
11115169841152 flags data|raid1
[ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage:
move data extents
[ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage:
update data pointers
[ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0

The problematic block group was balanced successfully this time.

I balanced a few more successfully (without the -dconvert=ngle option), then decided to reboot under 5.10 just to see if I would hit this issue again.
I didn't: the btrfs dev del worked correctly after the last 500G or so data
was moved out of the drive.

This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":

Overall:
Device size:        41.89TiB
Device allocated:   21.74TiB
Device unallocated: 20.14TiB
Device missing:      9.09TiB
Used:               21.71TiB
Free (estimated):   10.08TiB (min: 10.07TiB)
Data ratio:             2.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:       yes (data)

Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T     3.00GiB

Data,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta       7.14TiB
/dev/mapper/luks-10Tb       7.10TiB
missing                   482.00GiB
/dev/mapper/luks-new16T     6.95TiB

Metadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%) /dev/mapper/luks-10Tb      36.00GiB /dev/mapper/luks-ssd-mdata 36.00GiB

System,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta      32.00MiB
/dev/mapper/luks-10Tb      32.00MiB

Unallocated:
/dev/mapper/luks-10Ta       1.95TiB
/dev/mapper/luks-10Tb       1.96TiB
missing                     8.62TiB
/dev/mapper/luks-ssd-mdata 11.29GiB
/dev/mapper/luks-new16T     7.60TiB

I wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping
into the same issue on another FS, while rebalancing data after adding a drive,
which happens to be the old 10T drive of the FS above.

The btrfs fi usage of this second FS is as follows:

Overall:
Device size:        25.50TiB
Device allocated:   22.95TiB
Device unallocated:  2.55TiB
Device missing:        0.00B
Used:               22.36TiB
Free (estimated):    3.14TiB (min: 1.87TiB)
Data ratio:             1.00
Metadata ratio:         2.00
Global reserve:    512.00MiB (used: 0.00B)
Multiple profiles:        no

Data,single: Size:22.89TiB, Used:22.29TiB (97.40%) /dev/mapper/luks-12T        10.91TiB /dev/mapper/luks-3Ta         2.73TiB /dev/mapper/luks-3Tb         2.73TiB /dev/mapper/luks-10T         6.52TiB

Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%) /dev/mapper/luks-ssd-mdata2 32.00GiB /dev/mapper/luks-10T        32.00GiB

System,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb        32.00MiB
/dev/mapper/luks-10T        32.00MiB

Unallocated:
/dev/mapper/luks-12T        45.00MiB
/dev/mapper/luks-ssd-mdata2  4.00GiB
/dev/mapper/luks-3Ta         1.02MiB
/dev/mapper/luks-3Tb         2.97GiB
/dev/mapper/luks-10T         2.54TiB

I can reproduce the problem reliably:

# btrfs bal start -dvrange4625344765952..34625344765953 /tank
ERROR: error during balancing '/tank': No such file or directory
There may be more info in syslog - try dmesg | tail
[145979.563045] BTRFS info (device dm-10): balance: start
-dvrange4625344765952..34625344765953
[145979.585572] BTRFS info (device dm-10): relocating block group
34625344765952 flags data|raid1
[145990.396585] BTRFS info (device dm-10): found 167 extents, stage:
move data extents
[146002.236115] BTRFS info (device dm-10): balance: ended with status: -2

If anybody is interested in looking into this, this time I can leave the FS in this state.
The issue is reproducible, and I can live without completing the balance for the next weeks
or even months, as I don't think I'll need the currently unallocatable space soon.

I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).

I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave
like this, but on the first success, I won't know how to reproduce the issue a second time, as
I'm not 100% sure it can be done solely with the btrfs-image.

Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.html

Regards,

Stéphane Lesimple.

#5 lore raw
Quoted 6 lines
Quoted 5 lines

unfortunately the problem is no longer reproducible, probably due to writes happening in meantime. If you still want a btrfs-image, I can create one (unfortunately only without data as there is confidential data in it), but as the problem is currently no longer reproducible, I think it probably won't help.

Quoted 6 lines

That's fine, at least you get your fs back to normal.

I tried several small balance locally, not reproduced, thus I guess it may be related to certain tree layout.

Anyway, I'll wait for another small enough and reproducible report.

This is still reproducible on my FS, and I have the btrfs-image. I can easily upload it somewhere, but of course I understand downloading an image of 51G can be impractical.

An other way might be: as I know which block group is causing the problem, as per the dmesg, maybe I can dump only the part of the metadata relevant to this block group?

In any case I can run commands on this system, compile a custom btrfs-progs or a custom kernel with whatever you want me to try, and reboot as many times as necessary (this is not a production server).

I know it fails in relocate_block_group(), which returns -2, I'm currently adding a couple printk's here and there to try to pinpoint that better.

Regards,

Stéphane.

#6 lore raw
Quoted 2 lines

I know it fails in relocate_block_group(), which returns -2, I'm currently adding a couple printk's here and there to try to pinpoint that better.

Okay, so btrfs_relocate_block_group() starts with stage MOVE_DATA_EXTENTS, which completes successfully, as relocate_block_group() returns 0:

BTRFS info (device <unknown>): relocate_block_group: prepare_to_realocate = 0
BTRFS info (device <unknown>): relocate_block_group loop: progress = 1, btrfs_start_transaction = ok [...]
BTRFS info (device <unknown>): relocate_block_group loop: progress = 168, btrfs_start_transaction = ok BTRFS info (device <unknown>): relocate_block_group: returning err = 0
BTRFS info (device dm-10): stage = move data extents, relocate_block_group = 0
BTRFS info (device dm-10): found 167 extents, stage: move data extents

Then it proceeds to the UPDATE_DATA_PTRS stage and calls relocate_block_group() again. This time it'll fail at the 92th iteration of the loop:

BTRFS info (device <unknown>): relocate_block_group loop: progress = 92, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group loop: extents_found = 92, item_size(53) >= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>): add_data_references: btrfs_find_all_leafs = 0
BTRFS info (device <unknown>): add_data_references loop: read_tree_block ok
BTRFS info (device <unknown>): add_data_references loop: delete_v1_space_cache = -2
BTRFS info (device <unknown>): relocate_block_group loop: add_data_references = -2

Then the -ENOENT goes all the way up the call stack and aborts the balance.

So it fails in delete_v1_space_cache(), though it is worth noting that the FS we're talking about is actually using space_cache v2.

Does it help? Shall I dig deeper?

Regards,

Stéphane.

#7 lore raw
Quoted 14 lines · Stéphane Lesimple
Quoted 2 lines

I know it fails in relocate_block_group(), which returns -2, I'm currently adding a couple printk's here and there to try to pinpoint that better.

Okay, so btrfs_relocate_block_group() starts with stage MOVE_DATA_EXTENTS, which completes successfully, as relocate_block_group() returns 0:

BTRFS info (device <unknown>): relocate_block_group: prepare_to_realocate = 0
BTRFS info (device <unknown>): relocate_block_group loop: progress = 1, btrfs_start_transaction = ok [...]
BTRFS info (device <unknown>): relocate_block_group loop: progress = 168, btrfs_start_transaction = ok BTRFS info (device <unknown>): relocate_block_group: returning err = 0
BTRFS info (device dm-10): stage = move data extents, relocate_block_group = 0
BTRFS info (device dm-10): found 167 extents, stage: move data extents

Quoted 8 lines

Then it proceeds to the UPDATE_DATA_PTRS stage and calls relocate_block_group() again. This time it'll fail at the 92th iteration of the loop:

BTRFS info (device <unknown>): relocate_block_group loop: progress = 92, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group loop: extents_found = 92, item_size(53) >= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>): add_data_references: btrfs_find_all_leafs = 0
BTRFS info (device <unknown>): add_data_references loop: read_tree_block ok
BTRFS info (device <unknown>): add_data_references loop: delete_v1_space_cache = -2

Damn it, if we find no v1 space cache for the block group, it means we're fine to continue...

Quoted 6 lines

BTRFS info (device <unknown>): relocate_block_group loop: add_data_references = -2

Then the -ENOENT goes all the way up the call stack and aborts the balance.

So it fails in delete_v1_space_cache(), though it is worth noting that the FS we're talking about is actually using space_cache v2.

Space cache v2, no wonder no v1 space cache.

Quoted 2 lines

Does it help? Shall I dig deeper?

You're already at the point!

Mind me to craft a fix with your signed-off-by?

Thanks,
Qu

Quoted 5 lines

Regards,

Stéphane.

#8 lore raw
Quoted 51 lines · Qu Wenruo
Quoted 24 lines · Stéphane Lesimple
Quoted 3 lines

I know it fails in relocate_block_group(), which returns -2, I'm currently
adding a couple printk's here and there to try to pinpoint that better.

Okay, so btrfs_relocate_block_group() starts with stage MOVE_DATA_EXTENTS, which
completes successfully, as relocate_block_group() returns 0:

BTRFS info (device <unknown>): relocate_block_group:
prepare_to_realocate = 0
BTRFS info (device <unknown>): relocate_block_group loop: progress = 1, btrfs_start_transaction = ok
[...]
BTRFS info (device <unknown>): relocate_block_group loop: progress = 168, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group: returning err = 0 BTRFS info (device dm-10): stage = move data extents, relocate_block_group = 0
BTRFS info (device dm-10): found 167 extents, stage: move data extents

Then it proceeds to the UPDATE_DATA_PTRS stage and calls relocate_block_group()
again. This time it'll fail at the 92th iteration of the loop:

Quoted 10 lines

BTRFS info (device <unknown>): relocate_block_group loop: progress = 92, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group loop:
extents_found = 92, item_size(53) >= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>): add_data_references:
btrfs_find_all_leafs = 0
BTRFS info (device <unknown>): add_data_references loop:
read_tree_block ok
BTRFS info (device <unknown>): add_data_references loop:
delete_v1_space_cache = -2

Damn it, if we find no v1 space cache for the block group, it means we're fine to continue...

Quoted 9 lines

BTRFS info (device <unknown>): relocate_block_group loop:
add_data_references = -2

Then the -ENOENT goes all the way up the call stack and aborts the balance.

So it fails in delete_v1_space_cache(), though it is worth noting that the
FS we're talking about is actually using space_cache v2.

Quoted 8 lines

Space cache v2, no wonder no v1 space cache.

Quoted 2 lines

Does it help? Shall I dig deeper?

You're already at the point!

Mind me to craft a fix with your signed-off-by?

The problem is more complex than I thought, but still we at least have some workaround.

Firstly, this happens when an old fs get v2 space cache enabled, but still has v1 space cache left.

Newer v2 mount should cleanup v1 properly, but older kernel doesn't do the proper cleaning, thus left some v1 cache.

Then we call btrfs balance on such old fs, leading to the -ENOENT error. We can't ignore the error, as we have no way to relocate such left over v1 cache (normally we delete it completely, but with v2 cache, we can't).

So what I can do is only to add a warning message to the problem.

To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.

Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.

To verify there is no space cache v1 left, you can run the following command to verify:

# btrfs ins dump-tree -t root <device> | grep EXTENT_DATA

It should output nothing.

Then please try if you can balance all your data.

Thanks,
Qu

Quoted 9 lines

Thanks,
Qu

Quoted 5 lines

Regards,

Stéphane.

#9 lore raw

Hi,

Just for information: On my system the error appeared on a filesystem using space cache v1. I think my problem might then be unrelated to this one. If it will happen again, I will try to collect more information. Maybe a should try a clear_cache to ensure that the space cache is not wrong.

Bye,
David Arendt

Quoted 106 lines · Qu Wenruo
Quoted 51 lines · Qu Wenruo
Quoted 24 lines · Stéphane Lesimple
Quoted 3 lines

I know it fails in relocate_block_group(), which returns -2, I'm currently
adding a couple printk's here and there to try to pinpoint that better.

Okay, so btrfs_relocate_block_group() starts with stage MOVE_DATA_EXTENTS, which
completes successfully, as relocate_block_group() returns 0:

BTRFS info (device <unknown>): relocate_block_group:
prepare_to_realocate = 0
BTRFS info (device <unknown>): relocate_block_group loop: progress = 1, btrfs_start_transaction = ok
[...]
BTRFS info (device <unknown>): relocate_block_group loop: progress = 168, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group: returning err = 0 BTRFS info (device dm-10): stage = move data extents, relocate_block_group = 0
BTRFS info (device dm-10): found 167 extents, stage: move data extents

Then it proceeds to the UPDATE_DATA_PTRS stage and calls relocate_block_group()
again. This time it'll fail at the 92th iteration of the loop:

Quoted 10 lines

BTRFS info (device <unknown>): relocate_block_group loop: progress = 92, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group loop:
extents_found = 92, item_size(53) >= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>): add_data_references:
btrfs_find_all_leafs = 0
BTRFS info (device <unknown>): add_data_references loop:
read_tree_block ok
BTRFS info (device <unknown>): add_data_references loop:
delete_v1_space_cache = -2

Damn it, if we find no v1 space cache for the block group, it means we're fine to continue...

Quoted 9 lines

BTRFS info (device <unknown>): relocate_block_group loop:
add_data_references = -2

Then the -ENOENT goes all the way up the call stack and aborts the balance.

So it fails in delete_v1_space_cache(), though it is worth noting that the
FS we're talking about is actually using space_cache v2.

Quoted 8 lines

Space cache v2, no wonder no v1 space cache.

Quoted 2 lines

Does it help? Shall I dig deeper?

You're already at the point!

Mind me to craft a fix with your signed-off-by?

The problem is more complex than I thought, but still we at least have some workaround.

Firstly, this happens when an old fs get v2 space cache enabled, but still has v1 space cache left.

Newer v2 mount should cleanup v1 properly, but older kernel doesn't do the proper cleaning, thus left some v1 cache.

Then we call btrfs balance on such old fs, leading to the -ENOENT error. We can't ignore the error, as we have no way to relocate such left over v1 cache (normally we delete it completely, but with v2 cache, we can't).

So what I can do is only to add a warning message to the problem.

To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.

Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.

To verify there is no space cache v1 left, you can run the following command to verify:

# btrfs ins dump-tree -t root <device> | grep EXTENT_DATA

It should output nothing.

Then please try if you can balance all your data.

Thanks,
Qu

Quoted 9 lines

Thanks,
Qu

Quoted 5 lines

Regards,

Stéphane.

#10 lore raw
Quoted 3 lines · David Arendt

Hi,

Just for information: On my system the error appeared on a filesystem using space cache v1. I think my problem might then be unrelated to this one.

Then this is more interesting.

There are two locations which can return -ENOENT in delete_v1_space_cache():

- No file extent found
This means something is wrong in backref walk.
I don't believe it's even possible, or qgroup and balance will be completely broken.

- delete_block_group_cache() failed to grab the free space cache inode There is another possibility that, we have free space cache inode in commit root (where data relocation reads from), but it's not in our current root.
In that case, -ENOENT is safe to ignore.

I guess you may hit the 2nd case, as your next balance finishes without problem.

Quoted 1 line

If it will happen again, I will try to collect more information. Maybe a should try a clear_cache to ensure that the space cache is not wrong.

Clear_cache itself won't remove all the existing cache, it just remove the caches when the block group gets dirty.

Thus we use btrfs-check to remove free space cache completely.

Thanks,
Qu

Quoted 112 lines

Bye,
David Arendt

Quoted 106 lines · Qu Wenruo
Quoted 51 lines · Qu Wenruo
Quoted 24 lines · Stéphane Lesimple
Quoted 3 lines

I know it fails in relocate_block_group(), which returns -2, I'm currently
adding a couple printk's here and there to try to pinpoint that better.

Okay, so btrfs_relocate_block_group() starts with stage MOVE_DATA_EXTENTS, which
completes successfully, as relocate_block_group() returns 0:

BTRFS info (device <unknown>): relocate_block_group:
prepare_to_realocate = 0
BTRFS info (device <unknown>): relocate_block_group loop: progress = 1, btrfs_start_transaction = ok
[...]
BTRFS info (device <unknown>): relocate_block_group loop: progress = 168, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group: returning err = 0 BTRFS info (device dm-10): stage = move data extents, relocate_block_group = 0
BTRFS info (device dm-10): found 167 extents, stage: move data extents

Then it proceeds to the UPDATE_DATA_PTRS stage and calls relocate_block_group()
again. This time it'll fail at the 92th iteration of the loop:

Quoted 10 lines

BTRFS info (device <unknown>): relocate_block_group loop: progress = 92, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group loop:
extents_found = 92, item_size(53) >= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>): add_data_references:
btrfs_find_all_leafs = 0
BTRFS info (device <unknown>): add_data_references loop:
read_tree_block ok
BTRFS info (device <unknown>): add_data_references loop:
delete_v1_space_cache = -2

Damn it, if we find no v1 space cache for the block group, it means we're fine to continue...

Quoted 9 lines

BTRFS info (device <unknown>): relocate_block_group loop:
add_data_references = -2

Then the -ENOENT goes all the way up the call stack and aborts the balance.

So it fails in delete_v1_space_cache(), though it is worth noting that the
FS we're talking about is actually using space_cache v2.

Quoted 8 lines

Space cache v2, no wonder no v1 space cache.

Quoted 2 lines

Does it help? Shall I dig deeper?

You're already at the point!

Mind me to craft a fix with your signed-off-by?

The problem is more complex than I thought, but still we at least have some workaround.

Firstly, this happens when an old fs get v2 space cache enabled, but still has v1 space cache left.

Newer v2 mount should cleanup v1 properly, but older kernel doesn't do the proper cleaning, thus left some v1 cache.

Then we call btrfs balance on such old fs, leading to the -ENOENT error. We can't ignore the error, as we have no way to relocate such left over v1 cache (normally we delete it completely, but with v2 cache, we can't).

So what I can do is only to add a warning message to the problem.

To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.

Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.

To verify there is no space cache v1 left, you can run the following command to verify:

# btrfs ins dump-tree -t root <device> | grep EXTENT_DATA

It should output nothing.

Then please try if you can balance all your data.

Thanks,
Qu

Quoted 9 lines

Thanks,
Qu

Quoted 5 lines

Regards,

Stéphane.

#11 lore raw
In reply to Qu Wenruo
Quoted 1 line
Quoted 1 line

Mind me to craft a fix with your signed-off-by?

Sure!

Quoted 22 lines

The problem is more complex than I thought, but still we at least have some workaround.

Firstly, this happens when an old fs get v2 space cache enabled, but still has v1 space cache left.

Newer v2 mount should cleanup v1 properly, but older kernel doesn't do the proper cleaning, thus left some v1 cache.

Then we call btrfs balance on such old fs, leading to the -ENOENT error. We can't ignore the error, as we have no way to relocate such left over v1 cache (normally we delete it completely, but with v2 cache, we can't).

So what I can do is only to add a warning message to the problem.

To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.

Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.

Quoted 8 lines

To verify there is no space cache v1 left, you can run the following command to verify:

# btrfs ins dump-tree -t root <device> | grep EXTENT_DATA

It should output nothing.

Then please try if you can balance all your data.

Your analysis is correct, I do have v1 leftovers as I commented on the [PATCH] you've sent.

Now, fixing the FS:

# btrfs check --clear-space-cache v2 /dev/mapper/luks-tank-mdata
Opening filesystem to check...
Checking filesystem on /dev/mapper/luks-tank-mdata
UUID: 428b20da-dcb1-403e-b407-ba984fd07ebd
Clear free space cache v2
Segmentation fault

Wow, okay. That's unexpected.

# btrfs --version
btrfs-progs v5.9 

(gdb) r
Starting program: /usr/local/bin/btrfs check --clear-space-cache v2 /dev/mapper/luks-tank-mdata [Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1". Opening filesystem to check...
Checking filesystem on /dev/mapper/luks-tank-mdata
UUID: 428b20da-dcb1-403e-b407-ba984fd07ebd
Clear free space cache v2

Program received signal SIGSEGV, Segmentation fault.
balance_level (level=<optimized out>, path=0x555555649490, root=0x555555645da0, trans=<optimized out>) at kernel-shared/ctree.c:930
930 root_sub_used(root, right->len);
(gdb) bt
#0 balance_level (level=<optimized out>, path=0x555555649490, root=0x555555645da0, trans=<optimized out>) at kernel-shared/ctree.c:930
#1 btrfs_search_slot (trans=trans@entry=0x55555e8b4d30, root=root@entry=0x555555645da0, key=key@entry=0x7fffffffe000, p=p@entry=0x555555649490, ins_len=ins_len@entry=-1, cow=cow@entry=1)

    at kernel-shared/ctree.c:1320
#2  0x00005555555e3da7 in clear_free_space_tree (root=0x555555645da0, trans=0x55555e8b4d30) at kernel-shared/free-space-tree.c:1161
#3  btrfs_clear_free_space_tree (fs_info=<optimized out>) at kernel-shared/free-space-tree.c:1201
#4  0x000055555558cd5f in do_clear_free_space_cache (clear_version=clear_version@entry=2) at check/main.c:9872
#5  0x000055555559acce in cmd_check (cmd=0x555555638900 <cmd_struct_check>, argc=<optimized out>, argv=0x7fffffffe490) at check/main.c:10194
#6  0x000055555556ae88 in cmd_execute (argv=0x7fffffffe490, argc=4, cmd=0x555555638900 <cmd_struct_check>) at cmds/commands.h:125
#7  main (argc=4, argv=0x7fffffffe490) at btrfs.c:402

(gdb)

Can v1 leftovers provoke this?

The patch you've sent for btrfs-progs might fix my problem as I wouldn't need to remove space_cache v2 first, so I may not hit this bug, but if you're interested in looking into this one too, we might kill one bird with two stones!

I'm leaving my FS as is waiting for your reply,

Regards,

Stéphane.

#12 lore raw
In reply to Qu Wenruo

Qu Wenruo - 29.12.20, 01:44:07 CET:

Quoted 10 lines

So what I can do is only to add a warning message to the problem.

To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.

Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.

Quoted 6 lines

To verify there is no space cache v1 left, you can run the following command to verify:

# btrfs ins dump-tree -t root <device> | grep EXTENT_DATA

It should output nothing.

I have v1 space_cache stuff on filesystems which use v2 space_cache as well, so…

the fully working way to completely switch to spacecache_v2 for any BTRFS filesystem with space cache v1, is what you wrote above?

Or would it be more straight forward than that with a newer kernel?

Best,

Signature · 1 line
Martin